Journal of Cleaner Production 19 (2011) 946e957

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

Environmental performance and practice across sectors: methodology

and preliminary results
Don Goldstein a, *, Rachel Hilliard b,1, Valerie Parker b, 2
a
Department of Economics, Allegheny College, Meadville, PA 16335, USA
b
Centre for Innovation & Structural Change, NUI Galway, Ireland

a r t i c l e i n f o a b s t r a c t

Article history: This paper introduces a new methodology for measuring and modeling manufacturers’ environmental
Received 19 January 2010 performance and the managerial and technological practices that affect it. Facility level licensing data are
Received in revised form used to develop indicators based on sector-specific criteria but capable of being analyzed across sectors,
15 December 2010
at various levels of aggregation. This addresses the problem that environmental performance and
Accepted 17 December 2010
Available online 1 February 2011
determinants tend to be highly context-specific, while modeling and policy interests are often more
general. Using Integrated Pollution Control (IPC) information generated EU-wide, this approach should
be capable of cross-country extension. The methodology is tested on a sample of Irish facilities in three
Keywords:
Environmental performance
sectors during 1996e2004. Preliminary results show its usefulness in exploring the determinants of
Cleaner technology environmental performance at the sector and cross sector levels. Word count ¼ 10,049.
Environmental indicators Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction
In this paper, facility level data from the Irish Environmental
Protection Agency (EPA) are used to develop and test measures of
manufacturers’ environmental performance and the practices that
might affect it. Ireland was an early adopter of the EU’s Integrated
Pollution Control (IPC) licensing programme. From the mid-1990s
through 2004, companies were required to meet IPC’s stringent
pollution prevention standards and report detailed information on
air and water emissions, waste and resource usage, and relevant
management and technology practices. After, 2004, IPC was
superceded by IPPC, Integrated Pollution Prevention and Control
(European Union, 2005). This dramatic shift in regulatory regime
provides a natural laboratory in which to study the efficacy of
various approaches.
To do so, one needs measures of environmental performance
and its determinants. We use IPC information generated EU-wide
to develop and test a methodology for measuring and modeling
these relationships. This approach is thus of special interest
because it should be capable of cross-country extension. By starting
from facility level data and constructing measures scored on sector-
* Corresponding author. Tel.: þ1 814 332 3340.
E-mail addresses: dgoldste@allegheny.edu (D. Goldstein),
rachel.hilliard@nuigalway.ie (R. Hilliard), valerie.parker@nuigalway.ie (V. Parker).
1
Tel.: þ353 91 492932.
2
Tel.: þ353 91 495971.
based criteria, but capable of cross sector comparison, our approach
also addresses a recurring theme in the relevant literature (dis-
cussed below) e that environmental performance and its deter-
minants tend to be sector-specific, while modeling requirements
and policy interests are often more general.
With respect to environmental performance, our methodology
seeks a middle ground between single-indicator measures at a high
level of generality and those that achieve context-specificity by
focusing on individual impacts in single-sector analyses.
While our approach on environmental performance follows and
extends somewhat a large literature, as discussed in Section 2, with
respect to environmental practice we have had to break new
ground. There is no research of which we are aware that develops
detailed, quantitative representations of companies’ technological
and organizational actions that might affect environmental
performance. The major innovation reported here is to build such
representations from the kinds of information generated by EU-
wide licensing programmes.
Section 2 reviews the relevant literature, and ends with
a statement of our major research questions. Section 3 describes the
institutional setting and data sources for our study. In Sections 4
and 5, we introduce measures of environmental performance and
of management and technology practices. Section 6 presents some
preliminary empirical results in modeling the determinants of
environmental performance, exploring the potential of the
measurement methodology. Section 7 concludes with a discussion
of the results.

0959-6526/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jclepro.2010.12.012
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
2. Literature review
As noted above, we draw upon a large literature on the
measurement of environmental performance. Ultimately we are
interested in how economic activities affect specific aspects of
ecosystem quality (human and other organisms’ health, non-biotic
physical parameters). For example, King and Lenox (2002) proxy
environmental performance by means of toxicity weighted emis-
sions. But more typically, our information on these effects is
limited, complex, and ambiguous. Thus, following standard practice
(Dewulf and Van Langenhove, 2005), we focus instead on proxi-
mate environmental impacts in the form of emissions, waste, and
resource usage. (We do make use of technical and scholarly liter-
atures in determining which emissions are “most important”
within each sector, as described in Section 4).
For comparability purposes, researchers have emphasized the
importance of normalising such impacts. Typically this is accom-
plished by expressing impacts per ‘functional unit’ of output (MEPI,
2001, p. 29). Unfortunately, often information on output is not
available (as establishd by Duffy et al., 2003, with respect to the
Irish IPC licensees in our study). Depending on data availability, it is
sometimes possible to normalize by expressing impacts relative to
those predicted in an industry wide regression (King and Lenox,
2002). In our case, we have had to rely upon available data that
can proxy for output, as described in Section 4.
Another problem addressed in the literature is the tension
between studying performance at the industry or even firm specific
levels, where specific context mediates outcomes, versus at an
aggregated or even single-indicator level where comparisons over
time and/or across firms are facilitated in a more generalizable way
(GEMI, 1998; MEPI, 2001; Dewulf and Van Langenhove, 2005). One
important single-index approach is the Productive Efficiency (PE)
indicator (Tyteca, 1999). PE uses linear programming techniques on
sector data to define a ‘productive frontier’ with respect to
combinations of resource input use, desired salable output, and
undesired polluting output; each producer’s PE can be compared to
a maximum attainable PE frontier. Another important aggregate
indicator is the Jaggi-Freedman (JF) index (Jaggi and Freedman,
1992). The JF index uses the same kinds of variables as PE, but for
each producer an average of the variables is computed, which is
then normalised per unit of output and expressed as a ratio with
the best performer’s value in that variable. A more recent approach
is Fijal (2007), who uses measures of materials and energy flows to
construct a single environmental impact index for specific
technologies.
Various studies point out that despite the compactness of the
aggregate indicators and their ability to generate uni-dimensional
rankings among companies in a sector, valuable information may
be lost in combining disparate variables, in terms of the links
between specific producer actions and corresponding environ-
mental results (see for example Young and Rikhardsson, 1996, p.
116.) For this reason, another basic approach to environmental
performance measurement has been to disaggregate, studying
particular normalized impacts within specific industry settings. A
major, widely cited example is MEPI (2001).
There are also approaches that fall somewhere between single
measure indicators, with their high degree of generality, and the
highly context-specific measures focused on particular impacts and
sectors. In this vein, Dewulf and Van Langenhove (2005) create
slightly more disaggregated “environmental sustainability indica-
tors,” to be used in comparing alternative technologies within
a given sector. Also, Karavanas et al. (2009) create facility level
indicators that permit comparison with sector peers in either dis-
aggregated or single-indicator performance measures, and
comparison of sectors in the latter. They create “sub-indicators”
947
within each of a number of performance “components” (energy
use, emissions, etc.) and then choose the highest-impact sub-
indicators in aggregating up; we have employed a similar logic in
constructing the “key emissions” variable described in Section 4.
But Karavanas et al. do not attempt to analyze facilities’ perfor-
mance cross sectorally, a key goal of the present study.
While our methodology for environmental performance follows
and extends somewhat this large literature, with respect to envi-
ronmental practice we have had to break new ground. Dewulf and
Van Langenhove’s (2005) sustainability indicators are applied to
broad sector level technological alternatives (e.g., photovoltaic
solar versus natural gas-fired electricity-generation). But there is no
research of which we are aware that develops detailed, quantitative
representations of companies’ technological and organizational
actions that might affect environmental performance. The major
innovation reported here is to build such representations from the
kinds of information generated by EU-wide licensing programmes.
In this study we use such information, for IPC-licensed Irish
manufacturers, to create environmental performance and practice
variables and test their utility in analyzing the relationships
between the two. The basic questions to be addressed are thus the
following:
2.1. Research question one
Can we extend the literature by using IPC licensing information
to create environmental performance measures that permit flexi-
bility in moving between more and less aggregated indicators,
building up from facility and sector-specific considerations while
preserving broad comparability?
2.2. Research question two
Can we create corresponding measures of environmental prac-
tice, based on disaggregated facility level activities in the context of
sector level best practices, but capable of comparative aggregation
across industry sectors and practice categories?
2.3. Research question three
Are these measures useful in exploring the relationships
between environmental practice and performance of IPC-licensed
manufacturing facilities in a multi-sector setting?
Before addressing question one in Section 4, question two in
Section 5, and question three in Section 6, we describe the infor-
mation and manufacturing facilities used in the study.
3. Data sources and sample
3.1. IPC licensing in Ireland
Ireland’s Environmental Protection Agency Act introduced
Integrated Pollution Control licensing (IPC) of industry in 1994.
Formerly, firms complied with static emission limit values for air
and water, set at the time of licensing and not subject to subsequent
review. The IPC regulations, in contrast, included the following key
components:
3.1.1. Environmental technology
Standards for water and air emissions were set with regard to
BATNEEC (best available techniques not entailing excessive cost),
requiring all facilities to work towards attaining current BATNEEC.
The explicit aim is the development of environmental strategies
focused on cleaner technology, rather than ‘end of pipe’
948 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
approaches, making “waste minimization.a priority objective’
(EPA, 1996, p. 1).
3.1.2. Environmental management
Progress toward cleaner production was to be carefully planned,
managed, and reported. Licensed firms were required to develop
a five year environmental management programme of projects and
to submit an Annual Environmental Report (AER) to the EPA.
Included in the AER are details of all environmental projects being
carried out, with measurable goals, target dates and results. The
Irish EPA has been unusual among EU regulators3 in its explicit
focus on the activity content of structures for environmental
planning and management, including ‘document control, record-
keeping, corrective actions etc.’ (EPA, 1997, p. 7).
Facility information available at the EPA includes monitoring
results for specific emissions; reports of audit visits by the EPA
inspectors; correspondence between the firms and the Agency; and
the AERs. These sources provide detailed records of managerial
activities, technology projects, and environmental outcomes for the
years under license. In addition, separate license application files
contain information about technologies and systems in place,
providing a snapshot of pre-license period activity and expertise.
This combination of sources does much to alleviate problems of
‘green-glossing’ of self-reported information.4 The AERs are
prepared by managers or their hired consultants, and would be
expected to paint where possible a rosy picture of environmental
performance. But the abundance of EPA inspector reports and
Agency-generated correspondence in the files offers a check, and
provides information through which to filter claims by the regu-
lated entities.
3.2. Sample selection
We organize our study around three industry sectors. For
generalizability, we sought sectors exhibiting a range of technology,
product and market characteristics.5 The sample starts from all IPC-
licensed firms in each sector, defined by NACE categories, beginning
with companies sharing four digit NACE codes, but also chosen
from the three and even two digit levels when other information
suggests a company ought to be included:
3.2.1. Metal fabricating
NACE codes 2811, 2812, 2821, 2822, and 2840. Products include
electronics enclosures and cabinets; containers and tanks; struc-
tural steel and builders hardware; and radiators and heating panels.
Common processes are forging or pressing, cutting, welding,
degreasing and cleaning, and coating. Environmental impact-
reducing technologies include segregation and recycling of used
oils and waste metal, low-VOC or non-solvent cleaning and
degreasing, and water-borne, high-solids, or powder coatings. We
exclude facilities engaged predominantly in electroplating or
casting, because these are very different processes.
3.2.2. Paint and ink manufacturing
Primary NACE code 2430. Products may be solvent or water
based. Processes involve mixing of pigments and bases, either
3
A similar approach is taken in the Netherlands (Wätzold et al., 2001).
4
We thank an anonymous reviewer for this journal for pointing out this problem
in the present context.
5
Two additional factors facilitated linking the environmental data with financial
results for future analysis. First, we favored industries with a high percentage of
single-facility firms, where facility level environmental data would match with
company level financial data. Second, we avoided industries subject to substantial
transfer pricing bias due to facility ‘sales’ to same-company subsidiaries elsewhere.
manufactured on site or purchased. The key environmental concern
is VOC emission; thus water vs solvent based product is a key
variable. Manufacturing issues include (non-)enclosure of storage,
transfer, and mixing equipment; disposal vs separation and
recovery of wash water and/or solvents for equipment cleaning;
and handling of waste product.
3.2.3. Wood sawmilling and preservation
NACE codes 2010 and 2030. Processes involve cutting rough
wood to shape and size, and pressure treatment for water resis-
tance. Typical products are construction lumber, building frames
and roof trusses, posts, and fencing. Toxic pressure treatment
substances vs non-toxic alternatives is an important element in
environmental performance. We have excluded facilities making
composite products such as plywood, fiber board, or veneer
products.
The sample consists of 59 facilities with significant amounts of
data reported for the variables described below: 21 in metal
fabrication, 13 in paint and ink, and 25 in wood preservation and
products. The panel of data extends from 1996 (when IPC licensing
began for these companies) through 2004 (after which IPC
licensing was superseded by IPPC, Integrated Pollution Prevention
and Control). It is an unbalanced panel, as not all years (especially
early ones) are represented for all variables and firms.
We now turn to constructing measures of environmental
performance and practice for these facilities.
4. Measuring environmental performance
4.1. Research question one
Can we extend the literature by using IPC licensing information
to create environmental performance measures that permit flexi-
bility in moving between more and less aggregated indicators,
building up from facility and sector-specific considerations while
preserving broad comparability?
The EPA’s licensing approach has been that the performance
phenomena that matter, and how they are to be assessed, are highly
context-specific. Each facility has its own requirements with
respect to what environmental impacts must be monitored, how
often, and what limits are permitted. EPA files contain information
about licensed facilities’ impacts in terms of pollutant emissions,
generation and disposition of wastes, and resource usage.6 For each,
we construct common variables that are scored according to the
state of the knowledge on sector-specific environmental consider-
ations, but compared across sectors and over time to analyze a wide
range of possible relationships. The steps employed for all three
impact measures to create this comparability (with exceptions as
noted below) include normalization and within-sector averaging:
4.1.1. Normalising raw data
Measures of facility emissions, waste and resource usage must
be normalised by some standard unit of production scale, in order
to be meaningfully comparable over time and across firms. Most
impact data are in mass units e for example, kg/year. Ideally, these
data from facilities in the same sector could be normalised and
compared per ‘functional unit’ of output (MEPI, 2001, p. 29).
Because output data is not available for our sample, we normalize
6
The Correspondence in EPA files contains indirect data on environmental
performance: notifications of regulatory non-compliance involving actual envi-
ronmental impacts. We also construct a performance indicator based on this, which
is omitted in the following to conserve space.
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
mass impacts relative to a proxy, the number of employees (avail-
able in financial reports at the Companies Registration Office).7
4.1.2. Within-sector averaging
Once the raw data are normalised, we calculate each annual
facility environmental impact value as a ratio with its sector
average.8 When expressed this way, above versus below sector
average facilities can be compared across sectors, abstracting from
the fact that what might be considered “good” performance is
different for each sector.9
These comparability steps are applied to three basic sets of
environmental performance measures: emissions, waste, and
resource usage.
4.2. Key emissions
There are two additional steps in constructing the key emissions
variable: choosing the ‘key’ emissions, and averaging across them
to achieve a single emissions indicator that makes use of all rele-
vant data for each facility.
4.2.1. Choosing which emissions are ‘key’
We want to include those pollutants which are of greatest
environmental concern in the industry sector. The EPA indicates its
judgment on this for each license holder when it specifies which
emissions must be monitored and reported, for air, sewer (effluent),
and surface water discharges. Other industry sources have been
used as well in determining which emissions are key in each sector.
4.2.2. Averaging the emissions
The facility’s value each year is a simple average of its individual
emission amounts, each having been normalised and expressed as
a ratio with sector average as described above. The advantage of
this approach is that we utilize the available data on emissions the
EPA and other authorities consider important for each facility. The
disadvantage is that we compare companies using a measure
whose component parts are not uniform across all firms. Ulti-
mately, this approach adapts to and reflects the considerable
heterogeneity in monitoring and reporting across firms in each
sector.
The emissions considered to be ‘key,’ and included in the above
construction when reported for a particular facility-year, are the
following.
4.2.3. Metal fabricating
IDEM (2004) suggests the key emission for this sector is volatile
organic compounds (VOCs) to air. The US EPA (1995a) also targets
7
There is a potential bias from normalising by employment, in that technology
change over time might increase productivity. If output per worker rises, then all
else equal, so will environmental impacts. With mass impacts in the numerator and
employment in the denominator, then, the measure may be biased upward as
technology changes over time. But if one expects that changing technology will
reduce (normalised) impacts, then the measure tilts the scales against what is
expected. We conclude that normalising by employment will not bias the analysis
in a way warranting concern.
8
Data integrity is guarded at this point by removing extreme normalised values,
and then requiring that sector averages in each variable contain data from at least
three companies. Extreme values arise mostly from erratic numbers self-reported
in the AERs, and we have attempted to pre-exclude those that seem clearly to
reflect measurement error. Remaining outliers are screened using ‘outer fence’
values derived from inter-quartile range analysis.
9
We divide each year’s facility impact by the sector average for that impact
across all sampled years, not the sector average for that year alone. We do this
because many sector averages show distinct time trends, mostly downward; but we
want the company’s impact index to vary with its own performance without (in
that respect) reference to performance relative to its sector.
949
Table 1
Sector Averages: Disaggregated Emissions Performance (Except pH, kg/year, nor-
malised per employee).
Metals Paints Woods
pH performance .634 .289 No data
Carbon to air- mass 59.0 39.1
COD to water - mass 8.04 2.47
Suspended solids- mass .582 .473
Zinc to water- mass .119 NA
Averages are computed across all company-years. pH is expressed as deviation from
7.5 (absolute value). Extreme values are excluded using inter-quartile range
method.
this, and in addition wastewater emissions of solvents, acids, and
(for facilities that electroplate) heavy metals. The Irish EPA sets
VOCs to air and the pH of sewer emissions as frequent reporting
requirements (a third and a half of the facilities, respectively).
Chemical oxygen demand (COD), suspended solids, and zinc also
frequently appear in the sewer emission requirements. Thus the
key emissions performance variable in metal fabricating incorpo-
rates any of the following that are reported: VOCs (measured as
carbon) for air; and pH, COD, zinc, and suspended solids for water.10
4.2.4. Paints and inks
The greatest environmental impacts of this sector arise from the
emission of VOCs to air during use, not manufacturing, of the
product (ERI, 2004). Nevertheless, in paint manufacturing, VOCs are
released when solvent based raw materials, intermediate stages,
and end product are exposed to air (in mixing or transfer) or water
(particularly during clean-up). In addition, particulate matter con-
taining VOCs and heavy metals (when present in raw material) can
be released to air during grinding of pigments and to water during
clean-up. From among these emissions of concern, we define key
emissions for paint and ink facilities to incorporate the following
that are frequently reported in the IPC records: VOCs to air; and pH,
COD, zinc, and suspended solids in water.
4.2.5. Wood sawmilling and preservation
Unfortunately, licensed facilities in this sector during this time
period reported emissions only in “flow” units (e.g., concentration
in mg/l). While representative flow sampling could theoretically
provide already-normalised emissions measures, EPA advised that
this could not be assumed, and hence the wood sector facilities are
excluded from the emissions data and analysis.
The Key Emissions variable as constructed averages 1.0 within
each sector. To give a feel for the measurement system being used,
Table 1 backs up a step and shows how the sectors differ in the
elements underlying Key Emissions.
4.3. Waste
Waste is classified in the IPC facility documents as either
‘hazardous’ or ‘non-hazardous’ depending on the severity of its
potential impacts; and its ultimate handling is classified as
involving ‘recovery’ via some kind of treatment and reuse, versus
‘disposal’ to the environment (e.g., via incineration or land filling).
Combinations of these disaggregated variables have been used to
create three environmental performance variables in waste, each
10
For each of the emission components except pH, higher values indicate greater
impact and worse environmental performance. For meaningful inclusion with the
others, we set a pH value of 7.5 as the centre of an acceptable range for facility
wastewater (Hutchinson, 2008; Palmer, 2008), and define the corresponding
performance indicator as the absolute value of the difference between 7.5 and
a facility’s annual pH measure.
950 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957

Table 2 Table 3
Sector Averages: Waste Performance. Sector Averages: Resource Performance.

Metals Paints Woods Metals Paints Woods

Total waste (tonnes/employee) 7.97 6.05 173.19 Electricity end-use (MWh/employee) 12.36 10.02 42.32
Percent hazardous 14.8% 23.1% 5.7% Primary fuel (MWh/employee) 8.91 21.14 32.65
Percent disposed 32.8% 47.1% 40.3% Water (m3/employee) 78.76 80.71 22.11

Averages are computed across all company-years. Extreme values have been Averages are computed across all company-years. Extreme values have been
excluded using inter-quartile ranges. excluded using inter-quartile ranges.

expressed as ratios with their respective sector averages: Total

waste (normalised by employment), percentage of total waste that them together in a framework that is aggregationally flexible and
is disposed, and percentage of total waste that is hazardous (no based on data available in many EU settings. We turn now to less
normalization required for the latter two). Again, these final facility traveled ground, creation of similarly flexible and replicable envi-
variables expressed as ratios with their sector averages must ronmental practice measures.
average to 1.0 across each sector, and we provide a look at the
underlying sector waste differences by showing the sector averages 5. Measuring environmental practice
themselves in Table 2.
It is possible that the difference between the wood products 5.1. Research question two
sector and the other two in total waste reflects reporting errors or
inconsistencies. It is common for otherwise-wasted wood by- Can we create corresponding measures of environmental prac-
products to be collected and used as kiln fuel on site, and examining tice, based on disaggregated facility level activities in the context of
company records suggests that some may treat this as ‘waste’ while sector level best practices, but capable of comparative aggregation
others do not. On the other hand, the relatively low hazardous across industry sectors and practice categories?
waste percentage in woods suggests the progress made by these Firms take actions that may affect environmental performance,
facilities in substituting more benign preservatives for toxic ones, purposefully or otherwise, and we refer to such actions as ‘prac-
a suggestion that seems to be borne out in the statistical tests tices.’ IPC licensing resulted in documentation of an extraordinary
reported later. number and range of environmental practices. We distinguish
between practices involving technology and those characterized by
organizational systems or activities. We refer to the latter as
4.4. Resource usage
‘management practices.’
The EPA asks licensed facilities to report the annual use of
5.2. Management
electricity, fuel, and water in the AERs. We construct a variable for
each, again normalised relative to employment for comparability
There are three kinds of management practices that might affect
purposes, and expressed as a ratio to the relevant sector average for
environmental performance, by influencing the firm’s ability to
cross sector analysis. For fuel the variable we create is ‘primary
identify and act upon factors that can affect its environmental
fuels,’ which converts quantities of all fuel types consumed at the
impacts: planning, training, and procedural. We develop measures
facility to MWh equivalents (Carbon Trust, 2008)11 and sums MWh
of each by identifying and scoring discrete reported activities or
for each facility-year across fuel sources. (In the statistical tests in
projects of the appropriate type.
Section 6, we use for the resource performance variable ‘combined
resource use,’ the sum of the electricity, primary fuel, and water
5.2.1. Planning
variables described here.) What the energy variables measure is
This variable relates not to ‘planning’ qua orderly execution of
(the inverse of) energy efficiency; we assume that usage in MWh
pre-determined activities, but rather to processing of and/or search
provides a reasonable proxy for environmental impact (Carbon
for information in the course of evaluating possible courses of
Trust, 2008).12 Like for waste, above, we present the sector aver-
action. We use information from the AERs’ Environmental
ages themselves in Table 3 (Fig. 1).
Management Plans (EMPs) for ongoing and future pollution
Research question one has, we argue, been answered in the
reduction, and from the correspondence files, to construct a vari-
affirmative. We have used IPC-generated information to create
able to capture planning actions related to environmental perfor-
facility level environmental performance measures based on
mance. For each facility-year, the value of the management-
sector-specific criteria; normalization makes them comparable
planning variable is the sum of the year’s projects, scored according
over time and among industry peers, and sector averaging allows
to the degree to which concrete goals or targets are specified;
economy-wide comparability. Within each of the major areas e
relevant data or information is used to factor past experience
emissions, waste, and resource use e detailed impact measures can
systematically into decision making; and there is evidence of
by analyzed separately, or combined as aggregate indicators.
follow-through.
Specific techniques in this approach are drawn from the litera-
ture; our contribution vis-à-vis research question one is to bring
5.2.2. Training
By disseminating information about environmental impacts,
11 technologies, and/or management systems, employee training
Carbon Trust’s conversion factors, in kWh/m3 except as noted, are: natural gas,
10.9; diesel oil, 10,900; kerosene, 10,300; LPG, 7100; fuel oil, 11,900; and coal, programmes may affect companies’ environmental performance.
7472 kW h/tonne. We score training programmes according to their concreteness and
12
One could be more precise about at least one environmental impact, by esti- the extent to which they appear to drive changes in employee
mating tons of CO2 per facility from the electricity and fuel totals in MWh. In behavior. For each facility-year, the value of the management-
addition, a total energy efficiency variable could combine end-use electricity and
primary fuels. Both would require adjusting purchased electricity for national
training variable is the sum of the year’s projects, each scored
electricity-generation primary fuel mix and average transmission losses, and are according to the degree of specificity and the extent of change-
beyond the scope of the present study. creating follow-through.
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
Fig. 1. Summarizes the steps taken to construct the environmental performance variables.
5.2.3. Procedures
Tracking, recording, and reporting of regulated activities and
outcomes may affect environmental performance by providing
information on which impact-reducing steps can be based and
evaluated. We create and combine two components. One is the
timeliness and completeness with which EPA requirements are met
in the company’s AER (EPA, 1997): the EMP and data on the various
dimensions of emissions, waste, and resource usage. The other
component is EPA non-compliance notifications of a procedural
(rather than pollution-oriented) nature. The notifications use
a standard set of phrases to indicate the degree of severity assigned
to each non-compliance by the agency, which we use to create
a severity-weighted sum of the year’s procedural non-compliances.
The facility-year value for the management-procedural variable is
the sum of these AER (positive) and procedural non-compliance
(negative) scores.13
Table 4 summarizes the scoring criteria used. An example of
each variable, taken from Metals company 5 in 2001, may give a feel
for how they are constructed. The facility undertook a planning
project to assess means of segregating and recycling factory waste,
13
Thus the management procedures value can be negative. There is some
potential for double counting, as missing or incomplete AERs can generate notices
of procedural non-compliance. But investigation shows that EPA inspectors exercise
judgment in dealing with this kind of problem, and thus a non-compliance noti-
fication for inadequate AERs provides additional information beyond the AER
deficiencies themselves.
951
scored 2 out of a possible 3 because its goal was specific and it made
use of relevant data, but information on follow-through was lack-
ing; along with four other planning projects that year, this gener-
ated a 2001 management-planning value of 9. In that year the
facility implemented a (highly unusual) total of four training pro-
grammes, of which three received only one point due to a (very
common) lack of specificity and follow-through, creating a facility-
year management-training value of 6. Finally, the facility’s 2001
AER was scored a rather middling 5 points, offset by a whopping
negative 35 points of procedural non-compliance notices (one with
threat of ‘further enforcement,’ and eleven with the more serious
threat of ‘legal action’), for a management-procedural value of 30.
Table 4
Scoring Criteria: Management Practice Variables.
Planning Training Procedural
Execution (presence of project)
Specificity & concreteness
Use & quality of relevant data
Follow-through (extent of behavior change)
EPA sanctions (weighted by severity)
952 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
Table 5
Sector Averages: Management Practice.
Metals Paints Woods
Procedures 3.24 1.20 1.68
Planning 3.72 3.20 2.45
Training & Development 1.28 1.85 .51
Composite (Sum) 2.28 5.00 2.54
Table 5 shows sector averages for these management practice
variables, in addition to a combined variable formed from the sum of
the three. By two of three individual indicators and their composite,
the paint and ink manufacturing facilities appear to exhibit a higher
level of management practice. This impression is strengthened by
some of the statistical results reported in Section 6.
5.3. Technology
The license applications, AERs, and correspondence files contain
information about what we refer to as technology ‘projects’:
changes in the specific inputs, processes, and/or equipment by
which outputs are created. Documentation arises when facilities
seek EPA approval or advice on projects intended to reduce envi-
ronmental impact, or ones with potential environmental implica-
tions that are considered for other reasons. There are two main
challenges in transforming technology projects into appropriate
practice variables: defining the variables for cross sector analysis
while capturing sector-specific characteristics; and representing
the ongoing effects of prior years’ projects.
5.3.1. Cross-sector technology matrix
For each sector, we create a matrix within which technology
projects are located. One dimension of the matrix categorizes
projects according to a standard classification of pollution
prevention approaches. The other dimension of the matrix breaks
down each sector’s production process into major stages, according
to available technical sources on that sector. This matrix makes it
possible to test whether technological changes at particular points
in the production process, or using particular pollution prevention
approaches, are more or less important in improving environ-
mental performance.
The key feature of the technology matrix is that the stage of
production dimension is defined using sector-specific criteria, but
within a generalized schema common to all sectors. This allows us
to score technology projects using sector-specific criteria but
compare them across sectors in analyzing the data.
The pollution prevention dimension of the matrix uses the
following four categories (US EPA, 1995a,b):
- Raw materials e substitution with less polluting inputs,
elimination
- Closing the loop e segregation and on- or off-site reuse of
waste, product, and/or by-product
- Equipment changes e modification, replacement
- Process changes e not elsewhere counted
The stages of production dimension of the matrix uses five
general categories: product design, preparation, basic production,
finish work, housekeeping/other. Product design has to do with
basic characteristics and may involve choice of more or less envi-
ronmentally sensitive inputs; how the inputs are applied is scored
at the appropriate later stage. (E.g., in metals, choice of low-VOC
paint is scored in stage one; how the paint is applied is scored in
stage four.) These five general stages are specified as follows in
locating each sector’s projects:
5.3.1.1. Metal fabricating. Given the sample’s exclusion of facilities
whose primary activities are casting or electroplating, the stages
are (US EPA, 1995a; IDEM, 2004): 1) product design e especially
choice of the finish coating; 2) metal shaping e cutting, grinding,
forming, etc.; 3) surface preparation e cleaning, degreasing, etc.; 4)
finish coating e application of painting, plating, etc.; and 5)
housekeeping/other e storage, cleaning, bunding, waste handling,
packaging, etc.
5.3.1.2. Paints and inks. Typically pigments, base media, and other
materials are obtained from suppliers and then prepared and blended
at the facility (ERI, 2004; P2Rx, 2005): 1) formulation e choice of
base, pigments; 2) dry milling and mixing e drying raw materials
combined prior to wet processing; 3) wet milling and mixing e
further grinding and blending with wet materials; 4) filtering and
filling e final product preparation for shipping; and 5) housekeeping/
other e storage, cleaning, bunding, waste handling, etc.
5.3.1.3. Wood sawmilling and preservation. Rough logs are
prepared, cut, and treated for weather resistance (COFORD, 2004;
Environment Canada, 2002; US EPA, 1995b): 1) product design e
choice of pressure treatment chemical; also, sourcing lumber from
sustainably managed forests; 2) conditioning and cutting e
debarking, pre-drying, sawing; 3) treatment e impregnation of cut
wood with weather-proofing chemicals; 4) storage and drip-drying
of treated wood; and 5) housekeeping/other e storage, cleaning,
bunding, waste handling, packaging, etc.
Each project is assigned to the appropriate one among the 20
cells in the technology matrix for that facility-year e for example,
a raw materials substitution in the finishing stage of production.
The projects are scored on a scale of 1e5, depending on their nature
(end of pipe vs clean technology) and scope (how widespread
within that portion of the facility’s activity to which it could apply,
and/or fundamental in the production process). Clean technology
projects are those judged to prevent or reduce environmental
impacts (emissions, waste, and resource use) at the source; end of
pipe, in contrast, entails controlling a given impact once created
(Christie and Rolfe, 1995).
An example can illustrate the use of the technology matrix. In
1998, Metals facility 1 switched from a solvent based paint to a non-
solvent powder coating for most of its finished products. This is
a raw materials change in terms of pollution prevention approach,
at the product design stage. The project is assigned a score of 4 e
clean technology, applied to most but not all products e and this is
added to the total in the raw materials change e product design
stage cell of the matrix for that facility-year.
All project scores in each matrix cell for each facility-year are
added together. These disaggregated cells are combined as desired
to create the corresponding technology practice variables. In the
empirical work reported below, we have aggregated facility-year
cell totals across production stages, and alternatively across pollu-
tion prevention approaches. For example, we test the effectiveness
of loop closing projects at all production stages, or of projects at the
preparation stage across all pollution prevention approaches. The
algorithm for turning these project matrix cells into technology
practice variables has to do with impact over time, to which we
now turn.
5.3.2. Ongoing effects of prior years’ projects
Technology projects affect performance cumulatively over time.
But these effects decrease over time, as equipment depreciates, and
as the fit between projects and the surrounding production systems
in which they are embedded becomes less precise due to changes
elsewhere. A large literature suggests that technology investments
do not affect performance fully in the year of their implementation,
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957 953

Table 6 the most heavily used pollution prevention approach across the
Sector Averages: Technology Practice. sectors e a pattern that turns out to be of great interest regarding the
Metals Paints Woods practice-performance relationship, as shown in the next section e
Approaches Raw materials substitution 3.50 2.40 1.67 and the paints facilities’ composite advantage is maintained in
Closing the loop 2.41 4.33 2.59 equipment (Fig. 2).
Equipment investment 4.75 6.20 5.55 Research question two has, we believe, also been answered in
Process change NEC 2.60 2.05 3.04
the affirmative. These environmental practice measures are based
Stages Product design .46 2.06 1.59
Preparation 2.87 1.59 3.62 on fine-grained IPC information that, as in the environmental
Basic production 2.06 3.70 3.99 performance case, should be available in many EU settings. Sector-
Finish work 4.05 .64 1.42 specific criteria are used to score highly disaggregated measures of
Housekeeping/other 4.03 6.99 2.17 management practice in three categories and technology practice
Composite (Sum) 13.26 14.98 12.85
along the two dimensions of process stage and type of approach.
The shared structure of these rubrics makes management and
and that once fully operational the ‘efficiency schedule’ of invest- technology practice values comparable across sectors, and the
ment entails an approximately ten percent annual rate of decay in appropriate level of aggregation can be chosen to suit the analysis
impact (Doms, 1992). at hand.
While this literature deals primarily with fixed investment, in Together with the previous section’s results, this should allow
our data equipment projects represent less than half (about 40%) of the researcher to explore a wide range of empirical questions. The
the total. It is likely that there is less persistence in the effects of next section examines this expectation.
non-fixed technology projects. Therefore, we transform the sum-
med projects from the technology matrix cells into technology
practice variables assuming five year project lifetimes. Each new 6. Modelling the determinants of environmental
project’s score enters the variable at half its value in its first year, performance
full value the second, then 75, 50, and 25 percent of the original
value in project years three, four, and five. Thus for a given facility 6.1. Research question three
and year, for a particular technology matrix cell or combination of
cells e for example, all equipment-related projects, or all middle- Are these measures useful in exploring the relationships
stage projects e the corresponding technology practice variable between environmental practice and performance of IPC-licensed
reflects the cumulative influence of the active technology stock, manufacturing facilities in a multi-sector setting?
with the most recent projects (excepting the current year) There are at least two ways to test the usefulness of these
weighted heaviest and those more than five years old ignored. indicators in exploring the relationships between environmental
The weighted technology matrix approach as introduced here is performance and management and technology practices. One is to
designed so that technology practice variables are scored using look at individual companies’ impact-reduction efforts, and see the
sector-specific criteria, but the same set of variables is shared extent to which both the impacts and the efforts, measured
sample-wide for cross-sectoral analysis. The aim, as with the according to the indicators presented here, can be understood in
environmental performance and management practice variables ways that make sense in terms of what managers and regulators are
discussed earlier, is to facilitate both inference of cross-sector saying in each case. In Hilliard et al. (2010) we pursue this line of
dynamics and exploration of distinctions among the relationships inquiry, with results that answer research question three affirma-
at the sector level. Table 6 shows the sector averages, including tively. Another approach is to see whether broad, statistical prac-
a composite variable given by the sum of approaches or stages tice-performance relationships can be made sense of, using these
(either delivers the same total, summing across the rows or measures, on an ex-ante basis. That is the tack we take in the
columns of the weighted technology matrix). present study.
Table 6 shows that like in management, the paints sector has We begin with the three sectors combined, moving from highly
the highest total for technology practices. Equipment investment is aggregated to finer-grained practice measures. The following

Fig. 2. Summarizes the steps involved in constructing the management and technology practice variables.
954 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957

Table 7
Full-sample Partial Correlations: Environmental impact vs organizational practice (Probability values in parentheses).

Key emissionsd N ¼ 105 Total waste N ¼ 125 Combined resource use N ¼ 70

Technology (all categories, controlling for management & year) .196b (.047) .233a (.009) .419a (.000)
Management (all categories, controlling for technology & year) .179c (.071) .126 (.166) .045 (.745)

Based on Spearman’s rho.

a
Significant at 1% level (two-tailed).
b
5% (two-tailed).
c
10% (two-tailed).
d
Key emissions includes only metals and paints sectors facilities.

sub-section then explores differences and similarities among the
three sectors.
6.2. Cross-sectoral relationships
First we examine the relationship between aggregated practice
and performance variables. The management and technology
composite variables are used; on the environmental impact side,
we use key emissions, total waste, and composite resource usage
(electricity end-use, primary fuels, and water summed). The
statistics are Spearman’s rank-order correlations.14 Facilities’
management and technology practice values are themselves
correlated (Spearman’s correlation of .351, significant at 1%).
Therefore, in looking at the correlations between each kind of
practice and performance, we use ‘partial correlation’ to control for
the effects of the other practice e for example, the Spearman’s
correlation between emissions and management practice,
controlling for (holding constant, or removing) the effect of tech-
nology. We also control for the year in all tests, since many of the
variables exhibit time trends that may or may not be related to the
relationships of interest.
Table 7 shows the most aggregated statistical associations.15
While the most likely expected relationship would be negative,
we use a two-tailed significance standard to incorporate the
possibility of unanticipated positive relationships as well. Both the
management and technology practice composites are, as expected,
negatively correlated with emissions, at 10% and 5% levels of
statistical significance respectively. Combined management prac-
tice may be weakly correlated with reduced waste as well, although
not at an accepted level of statistical significance.16
An example of unanticipated directionality is the positive
correlation between technology and both total waste and
combined resource use. This unexpected result could reflect
reverse causality, with facilities that suffer from high waste levels
and/or resource usage undertaking technology investments inten-
ded to reduce them. Although there is a persistence effect built into
the technology variables, a sufficient lag in efficacy of these
investments could complicate inferences about the direction of
causality.
A comprehensive examination of the unexpected technology-
waste and impact association is beyond the scope of this paper; see
Hilliard et al. (2009a) for relevant results. But it is possible that
disaggregating the measures may shed some light by seeing if
certain kinds of technology projects are driving the unexpected
relationship. Since that will further explore the usefulness of the
measurement methodologies introduced in this study (research
question three), we turn there next.
We start with by disaggregating management practice into
procedures, planning, and training. To economize on degrees of
freedom, in each partial correlation of a disaggregated manage-
ment category we control for other management disaggregates
singly and for the aggregate technology variable, and vice versa.
Table 7’s negative relationship between emissions and
combined management categories is shown in Table 8 to be driven
by procedural and planning related management activities. Indeed,
training related management practice shows an unexpected posi-
tive partial correlation with emissions, and this extends to total
waste as well. It is possible that here as well, a kind of reverse
causality is in effect.
Tables 9 and 10 present corresponding partial correlations
between aggregate environmental impact measures and tech-
nology, disaggregated alternatively by approach to pollution
prevention and stage in the production process.
Emissions are negatively associated with two of the technology
approaches in Table 9, and with two of the technology stages in
Table 10; but there is a positive correlation with equipment
investment when breaking down technology by approach and with
basic processing when disaggregating by stage of production. As for
waste, we can now see that its unexpected positive correlation with
technology at the aggregate level (Table 7) appears to be driven
most consistently by a rather strong positive correlation with
equipment investments (Table 9). Resource usage, on the other
hand, shows a broad positive association with technology cate-
gories across the board.

Table 8
14 Full-sample Partial Correlations: Environmental impact vs management categoriesd
We have chosen nonparametric statistical techniques for the following reasons.
(Probability values in parentheses).
First, scatter plots of the data show that it does not conform even approximately to
the usual assumption of normal distributions. Related, many of the variables exhibit Key emissionse Total waste Combined resource
numerous extreme values, which can seriously bias parametric estimates. (We have N ¼ 105 N ¼ 125 use N ¼ 76
attempted to distinguish between measurement or recording errors, to be corrected
Procedure (controlling .232b (.020) .177c (.053) .058 (.629)
or excluded, and potentially legitimate values, of which we retain all but the most
for planning, training)
extreme as indicated by inter-quartile ranges.) In addition, it is difficult to specify
Planning (controlling .228b (.022) .110 (.231) .149 (.210)
a priori the functional form of many of the relationships of interest. Finally, we
for procedure, training)
cannot confidently attribute meaningfully uniform intervals to the values arising
Training (controlling .257a (.010) .185b (.042) .054 (.655)
from the data construction methods described in this study. Hence, we employ
for procedure, planning)
where appropriate analytical techniques based on rank-ordering.
15
The different numbers of observations in the columns of Table 7 reflect the Based on Spearman’s rho.
a
partial correlation calculation process, which begins with a set of simple correla- Significant at 1% level (two-tailed).
b
tions using only observations for which there is data on all three (here) variables of 5% (two-tailed).
concern e e.g., in column one, emissions, technology, and management. c
10% (two-tailed).
16 d
To conserve space, we test the determinants of total waste and combined Each partial correlation also controls for year and aggregate technology.
e
resource use only, and not the more detailed waste and resource indicators. 45. Key emissions includes only metals and paints sectors facilities.
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957 955

Table 9 value, the smaller sample sizes reduce their statistical significance.)
Full-sample Partial Correlations: Aggregate environmental impact vs technology by On the other hand, the full-sample association between combined
approachesd (Probability values in parentheses).
management practice and lower emissions is shown in Table 11 to
Key emissionse Total waste Combined resource be driven by the paints facilities alone.
N ¼ 105 N ¼ 125 use N ¼ 76 In addition, the sectoral breakout reveals more differences
Raw materials .037 (.713) .069 (.453) .230c (.054) regarding the puzzle of the positive technology-environmental
(controlling for others)
impact correlation in the full-sample results, for waste and resource
Closing loop .260a (.009) .068 (.459) .075 (.535)
(controlling for others) impacts. This positive association shows up most consistently in the
Equipment .199b (.047) .338a (.000) .259b(.029) wood products sector. The metal fabricating sector also exhibits the
(controlling for others) pattern found in the full-sample results, for technology vs resource
Process, NEC .357a (.000) .155c (.090) .227c (.057) use and perhaps (although not significantly) for total waste. But the
(controlling for others)
paint and ink sector does not display this positive correlation.
Based on Spearman’s rho.
a
A finer-grained look is provided in Table 12, which reports
Significant at 1% level (two-tailed).
b partial correlations for individual management and technology
5% (two-tailed).
c
10% (two-tailed). practice categories by sector. In the interest of space, we report only
d
Each partial correlation also controls for year and aggregate management. results that are at (P  10%) or near (P  20%) statistical signifi-
e
Key emissions includes only metals and paints sectors facilities. cance; full results are available from the authors. For ease of
interpretation, correlations at near-statistical significance are in
Table 10
Full-sample Partial Correlations: Aggregate environmental impact vs technology by parentheses, and positive ones (more practice e worse perfor-
stages (Probability values in parentheses). mance) are italicized.
Key emissionse Total waste Combined resource
In general, Table 120 s sector level disaggregation shows that
(N ¼ 105) (N ¼ 125) use (N ¼ 70) management practices appear to exhibit intended outcomes more
Product design .014 (.891) .101 (.275) .105 (.386) consistently in the paint and ink sector than in the other two. Even
(controlling for othersd) training and development, which for the full-sample correlates
Preparation .302a (.002) .033 (.720) .014 (.910) positively with emission and waste impacts, in paints is either
(controlling for othersd) associated with reduced impact (waste) or no effect. This may
Basic processing .167c (.099) .105 (.257) .410a (.000)
(controlling for othersd)
suggest that managers in the paints facilities are deploying training
Finish work .004 (.970) .223b (.015) .177 (.143) programmes proactively, not in reaction to environmental prob-
(controlling for othersd) lems as they arise. In metals and wood products, the results on
Housekeeping/other .232b (.021) .195 (.034) .301b (.011)
b
training continue to raise the question of whether a reverse
(controlling for othersd)
causality scenario, a lack of efficacy, or both are at work. On the
Based on Spearman’s rho. other hand, procedures and planning, where they exhibit signifi-
a
Significant at 1% level (two-tailed). cant values, tend to correlate negatively with the environmental
b
5% (two-tailed).
c
10% (two-tailed).
impacts they are intended to reduce.
d
Each partial correlation also controls for year and aggregate management. With respect to disaggregated technology practices at the sector
e
Key emissions includes only metals and paints sectors facilities. level, perhaps the most striking result in Table 12 is that in no
sector, and for no environmental impact measure, is equipment
We will discuss these findings further in what follows. But first, investment associated with improved environmental performance.
we explore whether additional insights can be generated by All of the statistically significant or near-significant correlations
examining these full-sample relationships on a sectoral level. between equipment investment and impacts are positive. The paint
sector is the only one not exhibiting this association, suggesting
6.3. Sector-specific relationships once more the possibility of a less reactive management dynamic.
Again, these sectoral results themselves leave open whether the
In Table 11 we look sector by sector at the broad correlations positive associations in metals and woods indicate unintended and
reported for the full-sample in Table 7, between environmental unwanted consequences, a reverse causality sequence, or some
impact types and combined technology and management practices. combination of the two.
First, the correlation between combined technology practice Other technology effects, both by pollution prevention approach
and lower emissions, reported for the full-sample in Table 7, is and by process stage, are mixed across sectors and environmental
shown in Table 11 to be shared between the metals and paints impact measures. There may be a weak indication that technological
facilities with usable emissions data. (While the sectoral tech- efforts focused at more fundamental change, by involving the
nology-emissions correlations are very close to the full-sample materials employed (raw materials approach), tend more to be

Table 11
Partial Correlations By Sector: Environmental impact vs combined organisational practice (Probability values, observations in parentheses).

Sectors Practices Impacts

Key emissions Total waste Combined resource use

Metal fabrication Technology (all categories) .193 (.142, N ¼ 61) .147 (.275, N ¼ 59) .455a (.013, N ¼ 31)
Management (all categories) .055 (.681, N ¼ 61) .051 (.704, N ¼ 59) .149 (.142, N ¼ 31)
Paint & ink manufacturing Technology (all categories) .182 (.249, N ¼ 44) .044 (.804, N ¼ 36) .125 (.589, N ¼ 23)
Management (all categories) .378a (.014, N ¼ 44) .115 (.516, N ¼ 36) .228 (.320, N ¼ 23)
Wood products & treatment Technology (all categories) No data .404a (.033, N ¼ 30) .510a (.021, N ¼ 22)
Management (all categories) .240 (.219, N ¼ 30) .053 (.823, N ¼ 22)

Based on Spearman’s rho.

Technology partials control for management, and vice versa; both control for year.
a
Significant at 5% level (two-tailed).
956 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957

Table 12
Partial Correlations By Sector: Environmental impact vs disaggregated organisational practice (Correlations with [10% < P values  20%] in parentheses; positive correlations in
italics).

Key emissions Total waste Combined resource use

Metal fabrication Management Management Management
Training .255c (Training .223) Planning .416b
Tech. approach Tech. approach Tech. approach
(Equipment .184) Equipment .302b None reportable
Process .376a Technology stage Technology stage
Technology stage None reportable 3: Basic process .434b
2: Preparation .470a N ¼ 59 5: Housekeeping .655a
N ¼ 61 N ¼ 31
Paint & ink Management Management Management
manufacturing Procedure .432a Training .303c (Procedure .309)
Planning .421b Tech. approach (Planning .332)
Tech. approach Raw materials .408b Tech. approach
None reportable Loop closing .420b None reportable
Technology stage Technology stage Technology stage
5: Housekeeping .387b 2: Preparation .383b 2: Preparation .518b
N ¼ 44 N ¼ 36 3: Basic process .672a
N ¼ 23
Wood products No data Management Management
& treatment Procedure .341c None reportable
(Training .317) Tech. approach
Tech. approach (Equipment .423)
Equipment .670a Technology stage
Technology stage (1: Product design .410)
4: Finish work .571a 4: Finish work .483c
N ¼ 30 N ¼ 22

Based on Spearman’s rho.

Technology partials control for other technology categories, combined management, and year.
Management partials control for other management categories, combined technology, and year.
a
Significant at 1% level (two-tailed).
b
5% (two-tailed).
c
10% (two-tailed).

associated with environmental impact-reduction. This result would
be consistent with a stylized fact reported frequently in the litera-
ture, that ‘cleaner technologies’ e those aimed at reducing impacts
at the source rather than cleaning them up at the ‘end of the pipe’ e
are most promising. In contrast, we note the concentration of
correlations between technological activity and greater environ-
mental impact in the later stages of production (Tables 10 and 12).
We believe these results confirm that the environmental perfor-
mance and practice measures introduced here are capable of sup-
porting useful, flexible empirical explorations, as asked in research
question three. Ex-ante reasonable expectations (better practice
improves performance) hold across much of the activity studied, and
the exceptions (reverse causality) are fairly systematic and can be
understood in equally reasonable ways. In the process, we learn much
about what drives behavior and what seems to work best.
7. Conclusion
This paper has introduced a new methodology for quantifying
facility level practices, both managerial and technological, that
might affect environmental performance. We have also built upon
the considerable existing literature on indicators of that environ-
mental performance. Our approach defines and measures practices
and performance based on sector characteristics, but in a way
permitting cross-sectoral comparison and analysis. This should
allow researchers, policy makers and managers to look for evidence
about what works at the very specific level, as well as for broader
regularities at various levels of aggregation. The methodology can
be implemented in any country whose environmental authority
has gathered the detailed information entailed in IPC (now IPPC)
licensing, and offers an opportunity to develop a rich empirical
representation of what companies are doing to address environ-
mental concerns.
Among the Irish manufacturing facilities studied, statistical
analysis of these indicators offers preliminary insights into the
practice-performance relationship. At a high level of aggregation,
organizational choices in management and technology are associ-
ated with improved performance with respect to the emissions that
are key for facilities in each sector. With respect to waste and
resource use, aggregate analysis reveals an unexpected association
between greater practice and higher impact.
Regarding both the expected and the unexpected broad
results, the methodology and data permit a finer-grained look at
individual sectors and at specific kinds of managerial and tech-
nology practice. This more detailed analysis suggests in
a preliminary way the presence of a reverse causality process,
whereby environmental impact problems stimulate increased
activity aimed at reducing those impacts; this seems especially
true for technology equipment investments in the metals and
woods sectors. Company level analysis (Hilliard et al., 2010)
suggests that in general, and particularly in these two sectors,
most managers avoid expensive steps like equipment purchases
unless pressed, in this case by regulatory scrutiny of severe
impacts. Ironically, facilities in the paints sector e with an
especially long history of regulatory pressure around VOCs and
related issues e do not exhibit the reverse causality phenom-
enon, having forged unusually strong managerial capabilities to
control the relationships between organizational effort and
reduced environmental impact.
Acknowledgements
The project was supported by the Environmental Protection
Agency of Ireland, the Center for the International Exchange of
Scholars and Irish Fulbright Commission, and Allegheny College.
We appreciate the many helpful suggestions received from
 D. Goldstein et al. / Journal of Cleaner Production 19 (2011) 946e957
participants at the 2007 conference of the U.S. Society for Ecological
Economics in New York City, and especially the assistance and
advice of participants at an October, 2007 Galway workshop on this
project: Frank Boons, Tim Bushnell, Andy Gouldson, Marcus Wag-
ner, and Kevin Woods. Two reviewers of our final report for the EPA,
and two referees for this journal, also made valuable suggestions.
Responsibility for all errors remains our own.
References
Carbon Trust, 2008. ‘Energy and Carbon Conversions.’ www.carbontrust.co.uk.
Christie, I., Rolfe, H., 1995. Cleaner Production in Industry. PSI Publishing, London.
COFORD, 2004. National Council of Forest Research and Development COFORD
Connects: Processing/Products. No. 4.
Dewulf, J., Van Langenhove, H., 2005. ‘Integrating industrial ecology principles into
a set of environmental sustainability indicators for technology assessment’.
Resources, Conservation and Recycling 43, 419e432.
Doms, M.E., 1992. ‘Estimating Capital efficiency schedules within production
Functions’. Centre for Economic Studies, 92e94. U.S. Bureau of the Census, CES.
Duffy, N., McCarthy, C., Zoehrer, M., 2003. Environmental Benchmarking for IPC
Industries. Synthesis Report. EPA, Wexford, ISBN 1-84095-103-6.
Environment Canada, 2002. Wood preservation’ in the green Lane. www.ec.gc.ca/
toxics/wood-bois/pubs/assesmentsection4_e.html#_Toc504929728.
Environmental Protection Agency (Ireland), 1996. BATNEEC Guidance Note for the
Chemical Sector.
Environmental Protection Agency (Ireland), 1997. IPC Guidance Note for Annual
Environmental Report. Ardcavan: EPA.
Environmental Protection Agency (U.S.), 1995a. Profile of the Fabricated Metal
Products Industry (Sector Notebook Project, September).
Environmental Protection Agency (U.S.), 1995b. Profile of the Lumber and Wood
Products Industry (Sector Notebooks series).
ERI: Environmental Road-mapping Initiative, 2004. Paints and Coatings: Impacts,
Risks and Regulations. National Centre for Manufacturing Sciences. http://ecm.
ncms.org/ERI/new/IRRpaintcoating.htm.
European Union, 2005. The IPPC Directive. http://europa.eu.int/comm/environment/
ippc.htm.
957
Fijal, T., 2007. ‘An environmental assessment method for cleaner production tech-
nologies’. Journal of Cleaner Production 15, 914e919.
GEMI., 1998. Measuring Environmental Performance: a Primer and Survey of
Metrics in Use. Global Environmental Management Initiative, Washington, D.C.
Hilliard, R., Goldstein, D., Parker, V., 2009. ‘Testing environmental performance,
practice, and capabilities,’ Chapter 4. Final Report. In: Environmental Tech-
nology, Dynamic Environmental Capabilities and Competitiveness. Centre for
Innovation & Structural Change (CISC), NUI Galway; and Environmental
Protection Agency, Wexford; Ireland.
Hilliard, R., Goldstein, D., Parker, V., 2010. ‘Evidence on Search Routines in Small and
Medium Sized Companies’. Centre for Innovation & Structural Change (CISC),
NUI, Galway, Ireland.
Hutchinson, J., 2008. Personal Communication. Department of Environmental
Protection, Pennsylvania.
Indiana Department of Environmental Management (IDEM), 2004. Sector specific
technology transfer Guide: metal products industry, March. www.in.gov/idem/
oppta/p2/assessments/.
Jaggi, B., Freedman, M., 1992. ‘An examination of the impact of pollution perfor-
mance on economic and market performance: pulp and paper firms’. Journal of
Business Finance and Accounting 19 (5), 697e713.
Karavanas, A., Chaloulakou, A., Spyrellis, N., 2009. ‘Evaluation of the implementation
of best available techniques in IPPC context: an environmental performance
indicators approach’. Journal of Cleaner Production 17 (4), 480e486.
King, A., Lenox, M., 2002. ‘Exploring the locus of profitable pollution reduction’.
Management Science 48 (2), 289e300.
MEPI., 2001. Measuring the Environmental Performance of Industry. Final Report.
European Commission, Brussels.
Palmer, J., 2008. Personal communication. Allegheny College Department of Envi-
ronmental Science.
P2Rx, 2005. The paint and coating manufacturing Topic Hub. http://www.p2rx.org/
topichubs/toc.cfm?hub¼28andsubsec¼7andnav¼7.
Tyteca, D., 1999. ‘Sustainability indicators at the firm level: pollution and resource
efficiency as a necessary condition toward sustainability’. Journal of Industrial
Ecology 2 (4), 61e77.
Wätzold, F., Büstman, A., Eames, M., Lulofs, K., Schuck, S., 2001. ‘EMAS and regu-
latory Relief in Europe: lessons from national experience’. European Environ-
ment 11 (1), 37e48.
Young, C.W., Rikhardsson, P.M., 1996. ‘Environmental performance indicators for
business’. Eco-Management and Auditing 3, 113e125.